> but there was a statistical attempt to attune for these effects
Can you please explain how the statistical attempt to adjust for latency of "hospitalization" at days 14-21 post-vaccination would have taken place? I believe almost all people hospitalized at days 14-21 would have been infected at times we don't expect the vaccine to be effective from the trial data-- so seeing efficacy near 0 against hospitalization at days 14-21 is indeed exactly what we'd expect.
Surely if the vaccine is preventing hospitalization right after the second dose, it prevented infection some time before that, right??
Edit: Observational data is always a mess. This looks particularly bad. Efficacy is higher in the old than in the young?? This is opposite of what we saw in the RCTs. Evidence of massive betas even right after vaccination when we expect efficacy to be 0?? This effectively says the vaccinated group is nothing like the unvaccinated group.
I believe estimate of infection date is by MOH data, which tries to estimate infection date in patients by clinical data. As for other effects, the idea is to "present the full range of reasonable scenarios and show how they affect the estimation of effectiveness."[0].
I'm not saying the analysis is right (I lack the credentials to say that) but I can tell you Israeli HMOs have been disappointed by 1st dose efficiency and have been vocal about it in media.
Soon there'll be a published preprint by Clalit alongside Miguel Hernán and Marc Lipsitch (so not some Israeli-only effort, there's external review), and that is likely to be definitive. For now, Clalit is keeping a 2 dose regiment and applying no pressure to change it.
Seriously.. I'm going to drop this now, but this data is a mess.
The methodological assumption of inferring beta from the first few days implies, looking at the infection data, that people receiving the vaccine are significantly more careful/less susceptible to the virus than those not.
But looking at hospitalization, it implies that they are WAY WAY WAY more susceptible.
last edit, for real: I looked at the preliminary Cialit findings. This is much more sane: a case control study. They find benefit-- about 33% efficacy-- post day 14, unlike the data you've been showing.
It's not blinded, so it is potentially wholly consistent with the Pfizer data if those vaccinated are taking 2-2.5x the risk of the control cases. To me, this seems pretty likely.
And if not, it is a different subpopulation--- we don't know what the data would look like for a similar demographic in the Pfizer study.
And, of course, I don't mean that we positively know that 1 dose is super effective in the real world for a longer stretch. But, it's something that's extremely likely to be shown in a blinded RCT based on the data we have so far.
I lack the ability to defend that analysis, but as I mentioned, it's not the only analysis showing reduced 1st dose efficiency. AFAICT, there's a consensus with Israeli HMO researches it's not as good as the Pfizer trials.
>I looked at the preliminary Cialit findings. This is much more sane: a case control study. They find benefit-- about 33% efficacy-- post day 14, unlike the data you've been showing.
I suspect you're talking about a different analysis then the preprint I'm talking about. IIRC, the preliminary analysis you're talking about was released earlier and without external verification.
Anyway, I don't see how your criteria can be falsified. If the data were to show 16% 1st dose efficiency, it would have been possible to claim the vaccinated are taking 4-5x risk. But how can you separate that from the possibility 1st dose is 4-5x less efficient than in the trials? Same thing for the reported 33% efficiency, it could be vaccinated are merely 1.25x less careful.
If so, there's a reasonable chance 1st dose isn't enough for Pfizer and by implication Moderna.
> Anyway, I don't see how your criteria can be falsified. If the data were to show 16% 1st dose efficiency, it can be claimed the vaccinated are taking 4-5x risk. But how can you separate that from the 1st dose being 4-5x less efficient than in the trials?
It can't be. This is why you can't really trust case control trials, especially when the participants are likely to behave differently between the groups. This is why, if you have a choice between a CCT and a RCT, you take the RCT data.
And we have a RCT, so...
And the difference between the RCT and CCT is in the direction we'd expect... and the magnitude isn't crazy high, so...
Yes, but there are also plenty of ways bias can creep into an RCT. Also remember we're not talking about the same virus - the trials were on 'vanilla' virus while by the time Pfizer arrived to Israel en mass B117 was prevalent. It could be the 1st dose is less efficient with B117. That's close to what the other link I provided earlier[0] argued. Can decision makers really take that risk?
> but there are also plenty of ways bias can creep into an RCT.
Yes, but in this case generally in the direction towards less efficacy too. People are unblinded inadvertently by their side effects and then act more recklessly, etc.
And this is immaterial: yes, RCTs can be fucked up, but CCTs are almost always fucked up.
> Also remember we're not talking about the same virus - the trials were on 'vanilla' virus while by the time Pfizer arrived to Israel en mass B117 was prevalent. It could be the 1st dose is less efficient with B117. That's close to what the other link I provided earlier[0] argued.
It could be. Anything could be. The question is the likelihood. But the other source arguing negative efficacy when even shitty case control data is showing positive efficacy is a little dubious.
Indeed, the hwole core assumption on that time series data is so bad-- assuming that people will behave the same from days 0-7 and from 7 on, when they've been cautioned that the vaccine is not immediately effective... Seriously, smh.
> Can decision makers really take that risk?
To do what I suggested? Yes. I suggested to run a trial. Don't pretend that I suggested anything different to strengthen your case.
It's a trial exceptionally likely to yield positive results for delaying the second dose.
There were suggestions earlier that 1st dose was "for sure" more effective, and judging by UK response some countries are desperate enough to try this without a trial. You didn't mean it, but it's possible some places will go for it. I'm just asking people to look before they leap, it's not a sure thing.
>Indeed, the whole core assumption on that time series data is so bad-- assuming that people will behave the same from days 0-7 and from 7 on, when they've been cautioned that the vaccine is not immediately effective... Seriously, smh.
Well, according to his github explanation[0] there was an attempt to account for that:
"It is assumed that on the days following the vaccination, there is increased caution to avoid social encounters."
>Well, according to his github explanation[0] there was an attempt to account for that:
> "It is assumed that on the days following the vaccination, there is increased caution to avoid social encounters."
Yes. He fudges beta by 0.25, when it has values >3. Uhh..
Edit: unless you actually propose that the vaccine is increasing risk of infection beyond changes in behavior, there's clearly something very big methodologically wrong here. And even that can't explain the drastically different betas assumed between the hospitalization series and the infection series. (I mean, they're on OPPOSITE SIDES OF 1 AND FAR AWAY FROM IT)
Well, you're trying my patience by putting words in my mouth that I didn't say: I didn't suggest decisionmakers immediately give everyone one dose. Are you making a dishonest argument, or did you just fail to read what I originally said.
I didn't mean to put anything in your mouth you did not say. If I accidentally offended you please accept my apology. Lets keep this to the science, as far as my limited ability allows me to engage, OK?
As for your link, according to the original PDF[0] this is raw data before any statistical corrections, which makes it not very useful. The person you're quoting has been noted as a relative pessimist, assumes that 1st dose effect is very small due to behavioural changes[1], and that a lot of the effect is due to the lockdown. (IMHO, he's too pessimistic, but again I'm far from an expert).
It's time series data. Yes, it's not case controlled. I said this already.
RCT > CCT > Time series.
So we've got RCTs for each Pfizer and Moderna that show similar first dose efficacy. We've got prelim CCT data that seems to show Pfizer efficacy but a bit lower than we'd like, confounded by possible behavioral changes. And we have time series data that seems to show first dose efficacy.
And then we have the very-confused-seeming over-tweaked analysis of time series data that you keep linking, too, comparing to base population rates without any effort at case control, which shows negative efficacy.
Anecdote: the people that I know who received their first dose decidedly did not become less active, either in the near term (actuality: slightly more active) or the longer term (actuality: considerably more active).
Assume that we do the RCTs etc. and discover the 'bit lower than we'd like efficiency' (33% vs the original 80%) the CCT you quoted argued, is the actual efficiency* , what then?
IMHO, Judging by what happened in Israel at the time (a rise in cases), I for one wouldn't recommend a one dose regimen. So I don't consider what I'm saying supported by only one analysis, but by all of the Israeli CCTs so far. Judging by the consensus of Israeli HMOs, they probably feel the same.
[EDIT: Basically, I don't consider a low efficiency all that different from '1st dose isn't efficient'. So we give people the 1st dose, they party, and we come up even.]
* Yes, people probably became more active, but remember there was a lockdown too, and a different variant than in the original RCTs. There's justification for a lowered efficiency beyond bias in the RCT.
> Assume that we do the RCTs etc. and discover the 'bit lower than we'd like efficiency' the CCT you quoted shows is the actual efficiency* , what then?
There's not just one figure of merit. The most important one is the ratio between (efficacy during the second month) with the delayed dose vs. with the on-time dose. If that's 0.5 or more, it's a slam dunk: you delay.
But post-dose 2 efficacy and immune titers and side effect profile matter, too. It's likely the long term efficacy is higher with delayed dose 2. This is what we see in most vaccines and 21 days is very, very early for a booster.
> remember there was a lockdown too,
Lockdown shouldn't affect anything in the CCT data. It's case controlled?? Presumably lockdown affects people in the two groups similarly, unless you're suggesting it reduces activity more in people who are vaccinated somehow..
>There's not just one figure of merit. The most important one is the ratio between (month 1 to month 2 efficacy) with the delayed dose vs. with the on-time dose. If that's 0.5 or more, it's a slam dunk: you delay.
So you're suggesting people with 1st dose keeping increasing the protection despite no booster in the delayed dose scenario. Possible, but important to point out that's not supported by the original RCT which takes no position on the matter.
>This is what we see in most vaccines and 21 days is very, very early for a booster.
True, but MRNA vaccines are new. 21 to 28 days could be the right period there.
>Presumably lockdown affects people in the two groups similarly, unless you're suggesting it reduces activity more in people who are vaccinated somehow..
No, but it dampens the 'people who get vaccine are partying' effect. I'm suggesting lockdown could prevent an effect you'd be seeing in the real world if we did this with no lockdown.
[EDIT: I'll respond to your other post here since this thread is already too sprawling.
>The only CCT preliminary data that I've seen suggests that one dose is protective. You have more CCTs?
My argument is that a low enough protection isn't so different from no protection. Lets say we give a delayed shot, people party for a longer time, the end result balances itself or could even be negative.
]
> So you're suggesting people with 1st dose keeping increasing the protection despite no booster in the delayed dose scenario. Possible, but important to point out that's not supported by the original RCT which takes no position on the matter.
No?
I'm suggesting we compare e.g. cases from day 21 to 42, when A) second dose is given on day 21, and B) second dose is given on day 42. (And people are blinded).
If efficacy B is at least half of efficacy A, then it makes sense to follow plan B.
Sorry I misunderstood. Your original phrasing implied a case when the protection increases (If month 1 / month 2 = 0.5 then month 2 is twice month 1).
Anyhow, it's important to note that the original RCT takes no position on what the result [month 1 / month 2] would be. It's a worthy thing to check, but it's not supported by the RCTs or CCTs.
>But one would expect that protection from days 21-41 without a second dose is comparable to protection from days 14-21.
I'm not sure that's enough for two reasons.
First, There's a study I linked to earlier, suggesting the 1st dose antibodies aren't good against the variants[0]. That would mean we need the booster ASAP.
Second, if the original efficiency is low, we're putting people at risk longer even if it stays about the same in 21-41. Imagine the 33% figure is correct. I don't think FDA would have approved the vaccine if efficiency < 50%, so should we keep people with such a low protection for another 20 days?
> First, There's a study I linked to earlier, suggesting the 1st dose antibodies aren't good against the variants.
It suggested the neutralization titers were worse, not that they were ineffective.
> so should we keep people with such a low protection for another 20 days?
We compare the ratio between people delayed and not delayed.
If a delayed person keeps, say, 50% efficacy, and the person with the second dose has 95% efficacy, giving two people 50% efficacy is better than giving one person 95% efficacy. That's why I suggested the specific figure of merit that I did.
(That is, imagine you produce 1 dose per 21 days, and there's two people in the world. You could give, in successive 21 day periods:
A 50% protection, B 50% protection, A 95% protection, B 95% protection.... (population efficacy 25%, 50%, 72.5%, 95%)
or you could give A 50% protection, A 95% protection, B 50% protection, C 95% protection). (population efficacy 25%, 47.5%, 72.5%, 95%).
>If a delayed person keeps, say, 50% efficacy, and the person with the second dose has 95% efficacy, giving two people 50% efficacy is better than giving one person 95% efficacy.
That logic breaks down when the divided efficacy is too low, even if it's formally half of the original. Say efficiency is 33% (and lets pretend the original is 66%). The people who got the 33% efficient vaccine go out and party, balancing out the effect. People think the vaccine doesn't help. Eventually nobody trusts your vaccine, and nobody shows up to get the second dose. I'm not sure 50% is high enough. Maybe it is, but just barely.
> See figure B here taken from the study. That's a large difference to my completely untrained eye.
So, LoQ means limit of quantification, aka the threshold below which things can't be measured in their assay technique. I know it's a log scale, but...
The study was barely able to detect the response to the baseline variant, so it doesn't take all that much of a drop to escape the study's quantification limit.
It's extremely misleading to plot the line all the way down to the bottom of the graph, when all you know is that it is (probably) less than the LoQ line. I say "probably" because this assumes the authors have accurately determined the LoQ, which is easy to mess up.
> That logic breaks down when the divided efficacy is too low, even if it's formally half of the original. Say efficiency is 33% (and lets pretend the original is 66%). The people who got the 33% efficient vaccine go out and party, balancing out the effect. People think the vaccine doesn't help. Eventually nobody trusts your vaccine, and nobody shows up to get the second dose. I'm not sure even 50% is high enough
Err.. Just to be clear, I never meant to suggesty ou'd take the ratio between month 1 and month 2. I tried to use parens to indicate the parameter that would be compared between the two groups (delayed and not delayed). Anything else wouldn't be an RCT, but just gawking at time series data...
> [EDIT: Basically, I don't consider a low efficiency all that different from '1st dose isn't efficient'. So we give people the 1st dose, they party, and we come up even.]
I'm a bit confused by what this means.
> So I don't consider what I'm saying supported by only one analysis, but by all of the Israeli CCTs so far.
The only CCT preliminary data that I've seen suggests that one dose is protective. You have more CCTs?
> I believe estimate of infection date is by MOH data, which tries to estimate infection date in patients by clinical data.
I'm reading the methodology PDF and there appears to be no effort to correct for infection date.
edit: Examining the MOH data-- the date of first positive test result is used for infections, and the date of hospitalization is used for hospitalization.
Can you please explain how the statistical attempt to adjust for latency of "hospitalization" at days 14-21 post-vaccination would have taken place? I believe almost all people hospitalized at days 14-21 would have been infected at times we don't expect the vaccine to be effective from the trial data-- so seeing efficacy near 0 against hospitalization at days 14-21 is indeed exactly what we'd expect.
Surely if the vaccine is preventing hospitalization right after the second dose, it prevented infection some time before that, right??
Edit: Observational data is always a mess. This looks particularly bad. Efficacy is higher in the old than in the young?? This is opposite of what we saw in the RCTs. Evidence of massive betas even right after vaccination when we expect efficacy to be 0?? This effectively says the vaccinated group is nothing like the unvaccinated group.